An official website of the United States government
Language is subject to a variety of pressures. Recent work has documented that many aspects of language structure have properties that appear to be shaped by biases for the efficient communication of semantic meaning. Other work has investigated the role of social pressures, whereby linguistic variants can acquire positive or negative evaluation based on who is perceived to be using them. While the influence of these two sets of biases on language change has been well documented, they have typically been treated separately, in distinct lines of research. We used a miniature language paradigm to test how these biases interact in language change. Specifically, we asked whether pressures to mark social meaning can lead linguistic systems to become less efficient at communicating semantic meaning. We exposed participants to a miniature language with uninformative constituent order and two dialects, one that employed case and one that did not. In the instructions, we socially biased participants toward users of the case dialect, users of the no-case dialect, or neither. Learners biased toward the no-case dialect dropped informative case, thus creating a linguistic system with high message uncertainty. They failed to compensate for this increased message uncertainty even after additional exposure to the novel language. Case was retained in all other conditions. These findings suggest that social biases not only interact with biases for efficient communication in language change but also can lead to linguistic systems that are less efficient at communicating semantic meaning.
more »
« less
Unsupervised Selection of Negative Examples for Grounded Language Learning
There has been substantial work in recent years on grounded language acquisition, in which a model is learned that relates linguistic constructs to the perceivable world. While powerful, this approach is frequently hindered by ambiguities and omissions found in natural language. One such omission is the lack of negative descriptions of objects. We describe an unsupervised system that learns visual classifiers associated with words, using semantic similarity to automatically choose negative examples from a corpus of perceptual and linguistic data. We evaluate the effectiveness of each stage as well as the system's performance on the overall learning task.
more »
« less
- Award ID(s):
- 1657469
- PAR ID:
- 10066405
- Date Published:
- Journal Name:
- Proceedings of the 32nd Conference on Artificial Intelligence (AAAI)
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Expectation-based theories of sentence processing posit that processing difficulty is determined by predictability in context. While predictability quantified via surprisal has gained empirical support, this representation-agnostic measure leaves open the question of how to best approximate the human comprehender's latent probability model. This article first describes an incremental left-corner parser that incorporates information about common linguistic abstractions such as syntactic categories, predicate-argument structure, and morphological rules as a computational-level model of sentence processing. The article then evaluates a variety of structural parsers and deep neural language models as cognitive models of sentence processing by comparing the predictive power of their surprisal estimates on self-paced reading, eye-tracking, and fMRI data collected during real-time language processing. The results show that surprisal estimates from the proposed left-corner processing model deliver comparable and often superior fits to self-paced reading and eye-tracking data when compared to those from neural language models trained on much more data. This may suggest that the strong linguistic generalizations made by the proposed processing model may help predict humanlike processing costs that manifest in latency-based measures, even when the amount of training data is limited. Additionally, experiments using Transformer-based language models sharing the same primary architecture and training data show a surprising negative correlation between parameter count and fit to self-paced reading and eye-tracking data. These findings suggest that large-scale neural language models are making weaker generalizations based on patterns of lexical items rather than stronger, more humanlike generalizations based on linguistic structure.more » « less
-
Abstract Statistical learning (SL) is the ability to detect and learn regularities from input and is foundational to language acquisition. Despite the dominant role of SL as a theoretical construct for language development, there is a lack of direct evidence supporting the shared neural substrates underlying language processing and SL. It is also not clear whether the similarities, if any, are related to linguistic processing, or statistical regularities in general. The current study tests whether the brain regions involved in natural language processing are similarly recruited during auditory, linguistic SL. Twenty-two adults performed an auditory linguistic SL task, an auditory nonlinguistic SL task, and a passive story listening task as their neural activation was monitored. Within the language network, the left posterior temporal gyrus showed sensitivity to embedded speech regularities during auditory, linguistic SL, but not auditory, nonlinguistic SL. Using a multivoxel pattern similarity analysis, we uncovered similarities between the neural representation of auditory, linguistic SL, and language processing within the left posterior temporal gyrus. No other brain regions showed similarities between linguistic SL and language comprehension, suggesting that a shared neurocomputational process for auditory SL and natural language processing within the left posterior temporal gyrus is specific to linguistic stimuli.more » « less
-
null (Ed.)We introduce The Benchmark of Linguistic Minimal Pairs (BLiMP), 1 a challenge set for evaluating the linguistic knowledge of language models (LMs) on major grammatical phenomena in English. BLiMP consists of 67 individual datasets, each containing 1,000 minimal pairs—that is, pairs of minimally different sentences that contrast in grammatical acceptability and isolate specific phenomenon in syntax, morphology, or semantics. We generate the data according to linguist-crafted grammar templates, and human aggregate agreement with the labels is 96.4%. We evaluate n-gram, LSTM, and Transformer (GPT-2 and Transformer-XL) LMs by observing whether they assign a higher probability to the acceptable sentence in each minimal pair. We find that state-of-the-art models identify morphological contrasts related to agreement reliably, but they struggle with some subtle semantic and syntactic phenomena, such as negative polarity items and extraction islands.more » « less
-
null (Ed.)Aims and objectives/purpose/research questions: Language brokering (LB) is an informal translation experience where bilinguals serve as linguistic and cultural intermediaries for family members. LB may have long-term socio-emotional and cognitive outcomes, yet little is known about its effects on executive functions (EFs). This study examines how first language (L1) proficiency and negative emotions tied to language brokering experiences affect EF performance on a Simon task (ST). Design/methodology/approach: Fifty-three Mexican American Spanish–English bilinguals with LB experience performed a ST, and reported their feelings towards LB for their mother. Data and analysis: Mean reaction times (RTs) and accuracy rates for correct ST trials were analyzed using linear mixed effects modeling, with trial type, proficiency and negative emotions tied to LB experience as factors and their interactions as additional predictors. Findings/conclusions: The L1 proficiency and negative emotions tied to brokering experiences have divergent, but combined effects on EF. Contrary to our hypotheses, low L1 proficiency predicted better performance and the smallest Simon effect was found for brokers with low L1 proficiency and low negative emotional brokering experiences. However, high L1 proficiency predicted better performance (smallest RTs) regardless of negative emotions tied to brokering experiences. Originality: This study takes a different perspective on the examination of individual differences among bilinguals, in which we examine how negative emotions tied to brokering experiences coupled with L1 proficiency relates to EF performance. Significance/implications: Our results provide support for the need to understand how individual differences in bilingual language experiences, such as L1 proficiency and negative emotions tied to LB, interact with performance on the ST.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
